Академический Документы
Профессиональный Документы
Культура Документы
r=-
fjSTRIJMENTATION
;llCAI
E~·~··
CONTROLLER TUNING
AND
CONTROL LOOP
PERFORMANCE
Is E c o N n E n 1T 1o N I
~
:.
A Primer
By David W. St. Clair
PUBLISHED BY:
STRAIGHT-LINE CONTROL COMPANY, INCORPORATED
,~,
.·~
BACKGROUND
•
I
•
THIS BOOKLET WAS ORIGINALLY ISSUED IN 1983 AS AN INTERNAL REPORT IN
THE DUPONT COMPANY TO HELP ENGINEERS AND TECHNICIANS, WHO HAVE
NO SPECIAL TRAINING IN FEEDBACK CONTROL, UNDERSTAND THE BASIC
CONSIDERATIONS AND LIMITATIONS. IT HANDILY BROKE ALL RECORDS AT
THE DUPONT COMPANY FOR NUMBER OF REQUESTED COPIES (OVER 1200)
WHEN ISSUED. THAT REPORT WAS RELEASED TO THE PUBLIC AND
.
I •
A Primer
By David . St. Clair
PUBLISHED BY:
IGHT-LINE CONTROL CO., INC.
'
,
,- -.··-
'
''
'
• ,---'
-~. -
' -- - ',, •
- ';,-
'
'
' '
'·'
' .
,'
-~
'
'
.,
First Pr~ntillii·~f'~l 990 · .
Second Priritil\f, April 1990 .
Third Printin~.~*P~r,.1:~?0
Forth Printing: Octciber l~O
Fifth Printing: May 1991 .
Sixth Printing: February 1993
SECOND EDITION
First Printing: December, 1993 '',_
'
• '•.
•
•
' 'i '
• '
?
' '
. ''-~- '
_:_~-
'
. . -·--
- •;.' ' '
.,,,.•
'
'
'
'
'.;_ .
'.
.
...,
••' '.
',. ' -
•'
.'.
'
.'
'
--'.'°! ,_,,
''
' ---- '
• .._.,,,
--~-
.. ,,
. . '" '
--
,' ' ;,
'
' '
'
•
. .
. ..
__ ---.'.
.. .. .
._.,.
•
.
. •
.
.. c·'.·-,,,,
:.:--~'.:o/f~~',
.
-' ..
PREFACE, v · ,.._-....
,._ ., . - :
- _,_; _-
.
.. .
. --~-''·'''···
._.,_ :-.- ' "
- -- - ··-
.-
' ,-
I
'l'ER 1, GETTING STARTED 1 - -.- '
'
.
•-- .· -
'
.
. · ·;;.,·c-:
-··
.
Science or Art . 2 • •
. Proportional t\ctjo1'- . 5 · ..
Integral Action .. ·. 8 .· -·· -
. .--'
-,
·- .._. :·--_·' . .
. '
- '
-.:·.;._
--·- - . -
•' '
•
-~--- ,_
Reset Windup 9 . . .
-·
'
,'
'•'•
'
-
''
,';
Y'
' C,
-•
.
,·
.,,,,
c'•i·
'
~:- •".
, .. ;
' --: :,. -
. -.
,.,,,._, ·-
;
.- .
. . '
-··-
-.
--
- - ._
.
'
... I
'
,,
-
-
--
1-
_,
• •• •..
;
- -'·_
~
•
. " -. -
• • • ••
·J
- '.·,
,.• -' . •• .!
-. -··'
. .
,._
_w
. .•,',.,_,
: '
,.,
----·>-~ :··-.
c
-'
Preparation •· 16 ··-·-
-- '' . - ;
-·--.,_;,·_:.
- - . ,_ '·' '
. .
. Closed-loop Tuning: What to Do 17 -· .. -- -----.
--,-
' - '
_,
-
·-.
-
•
. ..
·,
Closed-loop Tuning: How to Do It 18 -~-
' '-
' ·_, _ ; ' ' -- ··.·.·.'
"":"., -..
- .. __ ;
'
~ '
. -
-.- .
·--
.·'
'
.
·' _,_ --_ -- ,,
-
' '
. '
·-
. .
~--··---:.
-----' - .. -"
_ _
- -"
_,
_.,-- - -- -
'
Integrator 43 · -
, - , __·--"·- ------ ...
.- --- .
~~
,--,.-.
_____
_. ___
-·- .-- '
;
-
. -· .- c. - ' -. - • '•-.' • ; -- -_ - - - .
~
First Order Lag 43 -- . ' ' ' -.- ' '. --,_
__ __.. - .. - . ".
•
'
; :..
-
.
.-,: ·.--
45'
,.;_
Combining Building Blocks
Gains 47 .- -- - ~
--- -
'
. ',
'
- ' ' .
. '
•••
111
••
.
.
Dead Time •
Controllers ... ..
. ..
.
•
. . .
Tanks, Liquid Flow Lag 55
,• ·- • 'j. --
\• ,, --'
'
' - . ' -
... - ·'
·,·
• ' • '• :
.
. ,.,_
• ';
..
"· . ·~ ~
-··- -
,..
'.<'.'
•> •
-
.,._,- ' -
" --
--· . .
Interactions 65 ·_;.,•
''·-'
__ -
. .· ;.
:-
- - .-
. ....
Nonlinearities 66 .·
Process 67
Hardware, Continuous Nonlinearities 67 • "'; •
~
••
.
•
. •
' .
.
Valves at Limits 71 •
•
•
•
. -. - .._.
CHAPTER 9, POTPO ----- -
..
Digital Control Algorithms 73 ,_ \
.
. •. . '
'
s 81
•
GLOSSARY 83
APPENDIX A-1
Pure Dead Time Process A-2
Process with Dead Time and Integration A-3
Derivative Frequency Response A-4
Order For111 A-5
•
IV
•
PREFACE
This second edition of Controller Tuning and Control Loop Performance has
been extended in both directions frbm the first. Sections have been added for the
very beginner and for the somewhat more experienced. It is about twice the size.
Sections have been added on the what-to-do and how-to-do-it of tuning, to help
the person who may have never done it before. Then interspersed throughout are
paragraphs that extend some of the non-math concepts to the realm of math, or at
least algebra. These sections explaining concepts in math (sometimes frequency
response tenns) are clearly identified to make them easy to skip. This second
printing of the second edition also has expanded part of chapter 2 and has added
two pages to chapter 8, as compared with the first printing. It still stands on its
own, of explaining the essence of feedback control, without referring to math. I
hope these new references will help any reader who wants to bridge the gap from
the nonmath to the math.
The first edition was essentially a verbatim copy of a report written for DuPont in
1983, which I was allowed to make public in 1990. This second edition is perhaps
80% based on a 1992 update of that original report, written for a training course
for DuPont instrurnent technicians and engineers. The new version was to have
specific references to the DuPont situation, and was co-authored by Paul S.
Fruehauf (of DuPont) and myself (DuPont retired). I atn very appreciative of the
permission from Willia1n X. Alzos (of DuPont) to use what I wished from those
course notes.
I arn particularly grateful to Paul S. Fruehauf who has worn two hats in the
preparation of this second edition, first as co-author of the DuPont report, and
second as critical reviewer of my modifications and additions to that report. Most
of the material in chapter 2 is his. The first draft of much of that material was his,
and he persuaded me to include it in this booklet. He is cur1·ently an employee of
Applied Control Engineering, Inc., a consulting finn in Delaware.
I have tried to make this second edition appeal to readers whose background may
not be the chemical processing industries. I know I can only partly succeed in this
broadened scope, for all of my 40 years in the automatic control business were in
that industry.
I hope this booklet meets what I perceive as a need for more information on the
beginning end of training on the subject of controller tuning and control loop
perforn1ance.
ENJOY
This booklet on controller tuning and con- scapegoat, being blamed for problems that
trol loop perforrnance stops where most are not related to tuning, with the result that
books and courses on the subject begin. Too time and energy are spent needlessly.
often the subject is introduced with math Meanwhile a proper solution goes unsought.
unfamiliar to the reader. That does not have
to be there are simple concepts to help While I will give rules for tuning, the rules
those unschooled in the math to know and themselves are only~part of.the picture. The
understand the basics, to appreciate the ''tuner'' needs to know what the desired per-
limitations and to know what can be formance is and what to expect when the
expected. system is responding as well as can be ex-
pected, and when is it not. If it is not, then
the rules may not apply, or should be modi-
fied. This booklet teaches not only the rules,
but what can and cannot be expected
from tuning. It is also to teach some of the
common pitfalls. Why do the tuning rules
My field for 40 years was industrial process not seem to work sometimes? In addition,
control. The basic concepts of control are tuning is often done to fix some problem.
the same, regardless of the field. The exam- You cannot use or fix anything unless you
ples will change, but the concepts, princi- know how it should work, and that includes
ples, and much of the vocabulary won't. For control loops.
readers whose field is different from mine I
hope you will gain some useful insight into
your situation.
bearing on quality or other business consid- made? Not many. Quite possibly not any.
erations, so whether they are tuned tightly Usually there are at least a few loops that
or not is not all that important. How many stay on manual for some time, sometimes
new operations are started up and have all even years. It is hard to argue that these
the loops on automatic for the first product loops need tight tuning.
'
Controller tuning is mostly science. Tuning mathematically pure and simple models are
rules are based on mathematically clean and used to represent the ''typical'' process.
simple models that · approximate . the real Don't worry about that, certainly not at this
stage. The differences are relatively small
compared with what I consider realistic
goals in tuning. We w,ill not be concerned
about deter111ining settings to within 1%,
world. If .the . real world
•
were and generally not within 10 or 20%. For
mathematically clean . and simple, , then instance, if the tuning rules determine that a
controller tuning would be all science controller setting should be 1.00, it doesn't
(provided of course, there was agreement on really matter if it is set for 1.01 or 1.10.
what was desired from the tuning). Happily,
experience (and higher math) has shown
that the real world can be simplified without
sacrificing accuracy enough to worry about.
It is known, with a reasonable degree of
certainty, when this simplification is
invalid, and therefore when the rules for Even if set for 1.20 you would be hard
tuning will break down. pressed to see the difference in most
practical cases. Determining settings within
30 to 50°/o is a more realistic expectation.
Two specialists in tuning will almost surely
come up with different settings in any given
situation. They are far more likely to, indeed
will almost certainly, come up with the
same analysis of what may be wrong with a
There are ·numerous publications giving
loop. They are less likely to agree on what
tuning rules, and, as you might expect, they
the best solutio11 is. It is rather like politics
don't all give exactly the same rules. This is
because different. criteria are used for what in that regard.
constitutes ''proper'' tuning. Different
'
- -."• -
' ' - .- - '
No reasonably thorough writing on control- loop method. Nichols then verified the
ler tuning would be complete without pay- mathematical validity of the open-loop ap-
ing tribute to J. G. Ziegler and N. B. Nichols proach.
(Optimum Settings for Automatic
Controllers, Transactions .of the ASME, v For history buffs there is a book you should
64, Nov. 1942, p759). ·.Their contribution know about: Automatic Control, Gassical
was a quantum leap forward in the science Linear Theory, edited by George J. Thaler,
and/or art of tuning industrial controllers. It
,
Naval Postgraduate School. It was published
took perhaps.10 years or more after that be- by Dowden, Hutchinson and Ross Inc.,
fore subsequent authors started to hone and Stroudsburg, . !!A iJ:l .l 974. It ~s. out of print
- - .--.·-.--· . 7.' ,.. . .·
refine their recommendations, but the es- now, but.can be obu.ined from major techni-
sence of their approach .has remained un- cal libraries. The · book is orte of the
''Benchmark Papers in. Electrical Engineer-
'
' Nichols, with a mathematical bent, was pri- included in a collection of papers on PID
''
'
'
marily responsible for verifying the math of tuning: Reference Guide to PID Tuning,
the closed-loop formulas, while Ziegler, of a published by Control Engineering.
more empirical bent, conceived the open-
!
'
'
'
-
The task of tuning a controller can run from The science of control is based on math that
fairly simple to quite complex. It is rather is fonnidable to most persons. Happily it is
like income taxes. Many cases · are quite not necessary to understand or even use the
simple, but. then there· are a few that need a supporting math to absorb the governing
specialist. Also like taxes, it has to be done. principles. The math will be largely, if not
For the simpler cases, which constitute totally, omitted. No proofs will be provided.
possibly 80% (somewhere between half and There are however concepts, which may be
all) of the loops typically encountered, the new, which should be mastered. These
procedure can be reduced to a set of easy- relate primarily to understanding the
to-follow rules. importance TIME has in the control loop.
AMOUNT is also important, but not as
much as time. By far the most important
concept to master in understanding control
loops is the concept of LAGS. An effect
happens AFTER some cause. A control
valve moves AFTER the controller output
changes. The measurement of a temperature
While these rules are based on sound sci- in a well LAGS the actual temperature
ence, applying them without knowing what outside the well. The coldest day of the year
is expected leaves little understanding of occurs AFI'ER the shortest day of the year.
what is going on. Each tuning experience
becomes an isolated event. There is no
framework on which to build understanding,
no adequate way to transfer experience from
one time to the next, or from one person to
another. One goal of this booklet, is to pro-
vide that framework, that way of defining Not all lags are the same, or have the same
the experience so it is both understandable importance in a control loop. It will be a
to the person doing the tuning, and transfer- major part of this training material to de-
able to others. It might be called THE velop an understanding of where lags come
LANGUAGE OF CONTROL. Inciden- from, the different types that are used to ap-
tally, while I assume the reader has some proximate the real world, and what their
familiarity with many of the te11ns used in relative importance is. A few words, mean-
automatic feedback control, I have provided ing specific things in a control loop, will be
a glossary for the tertns most likely to need added to your vocabulary. Again, it is THE
defining. LANGUAGE OF CONTROL.
'
As with, I suppose, all fields, the science are sometimes available if the tuning is
and art of feedback control grew before the done with a digital control system. This
committees on standard ten11inology were constancy of the PID function in an era of
formed. The predicable result is that several
different terms are used to name the same
thing. The controllers we are going to talk
about have three adjustments. They are
froportional, Integral and Derivative (PID).
Computer based systems often have the
fourth, which is an adjustment for the filter phenomenal techni~al progress is a sobering
time. Computer based controllers may also thought. THE CIPLES HAVE
have a decision to be made about the cycle REMAINED UNCHANGED. I understand
time (how often the controller looks at the that some really high powered math has
process), but this is not considered a con- shown that the PID function is the best gen-
troller setting. It is, however, quite impor- eral purpose function to use to do the job.
I
[
tant, but it will be discussed in chapter 9 More sophisticated control algorithms will
!'' produce better performance when fitted to a
f PID controllers have been around since specific process, but poorer perfo11nance
'
! about 1940. Modem controllers perform the results if the process changes. This sensitiv-
i same functions as those, perhaps with a few ity to process changes is called robustness,
I
i
embellishments and certainly more accu- with more robust being less sensitive. The
'' rately, but the same functions nonetheless. PID algorithm is an excellent trade-off
i
So the tuning rules have remained essen- between robustness and perfor111ance.
tially the same over the years, though aids
•
6 Chapter 1, Getting Started
be an
offset if the contit>Jler has no integral '
'
'
'
0 (integral) action, to be described soon, ·
. Error, Percent of Scale then it will usually have a manual reset '
'
Figure 1.1. A proportional-only controller has a fixed (integral) adjustment.., This is an ·adjust-
relationship between error and output ment that allows som! manual compen-
sation for the offset.
direct acting. If it decreases as the con-
trolled variable increases, then it is called
reverse acting. The controller action is set
Controller
(or checked) initially, when a controller is Setpoint . Output
first put into service, and is not changed p
after that. The action has to be right to get
the controller . output to go in the right . - +
direction when the controlled variable Controlled Manual
·.-
I . - . .. .. '
.. ' '
. . ·- '
or more simply:
Kc
This is the first section dealing with
the math I algebra of control. If you since the lefl hand side of the
want to. skip. it, or any subsequent equation is understood to . be · the
section so devoted, I encourage you output divided by. the input., The
to do so. T/Jere is no need to get transfer function of any element in .• . .
bogged down in this and frightened the control loop is the ovtput divided ...
off what I am trying to say. My intent In.
by the input... this case it is what · · .
is to make the booklet stand alone you multiply the error by to get the
·without reference to math. These output: · ··. · ··. · •· .. · ··· ·· · ·
..
• •
brief sections · are presented to
introduce the math to •those· of you . · Once yeu get ihte the algebra you
. who might be interested,. with the ·.·.· can · satisfy yourself that these are ··.
hope that you won't be intimidated if ........ . legitimate simplifications. You should
you decide to read other material on ·• ·.· . not. play around·. vet}( much with the
the subject. ·•··. · .• .· algebra without learning a great deal
-_. ,.,_, ..
more than this booklet will teach you.
The · · math of · .• a · pfuportional-oniy · I have no intention of teaching you
·controller is quite simple: the rigors of the underlying math.
If you decide to learn more, you need
Output = (Error x Gain) + Bias to use another source.
8 Chapter 1, Getting Started
In the earlier days of industrial automatic was due to proportional action alone.
control the integral function was almost Within the physical constraints of the
universally called reset. Now the more sci- controller, the output will continue to
entifically correct te1·1n integral is gaining change at the same rate. This change comes
widespread use. I tend to use them inter- from integrating the error.
changeably, especially when talking as
compared with writing. When referring to So, the integral action causes the· controller
the adjustment the ter111s reset time, and output to change at a RATE proportional
reset rate are both in common use. One is
the reciprocal of the other, so of course it is
vital to know which one you are talking
about. To say to ''tum the reset up'' is an ...
ambiguous statement, because you don't =ea.
GI -
:s
I-iA
know whether the speaker is talking about --
c :s - - - __,,_I_ - -
reset time or reset rate. It usually means to ao I A
decrease the integral time, but the phrase __ l___t __
still leaves uncertainty. It is rather like I ....__Integral_ I
saying to tum the air conditioner up. Does I Time I
that mean to get more cooling or to tum the
ther·mostat higher? I will use reset time or
integral time when referring to the setting
...
itself, discouraging the use of reset rate. e...
wo.--
Integral action is not as easy to understand
as proportional action. The graph that is Time
often used to explain it is given as Figure Figure 1.4. A proportional-plus integral controller will
1.4, which is really for proportional-plus- integrate the error to add an amount to the output equal
integral action. Imagine a controller just by to the proportional change in one integral time.
itself, not connected to a process. Then to the error. The longer the integral time
imagine that from an initial condition for the slower it changes. A controller with
which the error is zero, that an error is integral actioh will eventually reduce the
suddenly introduced, called a step change. error to zero, as the output will continue to
The controller output will then change to a change until there is no error. That is, this
new value, and the amount of the change is will happen if there are no continuing
arbitrarily called ''A'' in Figure 1.4. disturbances to require the output to
continue to change, and if the manipulated
After that the controller output continues to variable has enough ''muscle'' to achieve
move in the same direction it went initially. that. Manufacturers build their integrating
It will move an amount equal to the initial function to be as close to mathematically
amount ''A'' in a time that is the integral pure as they can, and they do a good job of
time or the reset time. The units of reset it, whether it be one of the very first
time or integral time are minutes per repeat. pneumatic controllers, or one of the latest
The reason for this terminology is illustrated digital controllers.
in Figure 1.4, which shows that the integral
time is the time to repeat the change that
•
•
Before the advent of digital controllers there Not much more will be said about the reset
were integral-only c()ritrollers, but they were windup problem at this point, except to say
not in widespread use. The function is the two things. One is that it is a phenomenon
same as in a proportional-plus-integral
. . . •,.
that does exist, and two is that the measures
controller,· except of course there is no taken to combat the problem work with only
change in controller output due to varying degrees of success. These measures
proportional action. The change in con- seldom totally eliminate the problem. It is
troller output is all from integrating the er- far better to take steps to see that the
ror. With essentially all digital controllers controller does not windup in the first place,
there is the option to have integral-only than to expect the anti-windup features to
action. When this might be used will be keep you out of trouble. For batch processes
discussed later. reset windup can be an especially severe
problem on start up..· Specially configured
controllers exist to combat this problem, but
they will not be discussed in this booklet.
Any loop will cycle if you reduce the The math (algebra) for a propor-
integral time far enough. This is true tional-only controller had nothing in it
whether the controller is proportional-plus- relative to time. The proportional-
integral or only integral. The task of.setting plus-integra/ controller does. This
introduces a new symbol, which is
the integral time is one of setting it low
used in essentially all of the literature
enough but not too low. today, and that is the lower case ''s. ''
d
s=-
dt
Any control loop with integral action is If you didn't know what it was before,
subject to having a problem called reset you still don't! The d is the derivative
windup, or more recently, integral windup. dt
This refers to the condition when the con- relative to time. If you see !_, this is the
troller output does not have enough muscle s
reciprocal of derivative, which is
to reduce the error to zero. Since the con-
integral. Please simply accept that. The
troller integrates this error, the output will Laplace transform (transfer function) for
continue to change until it reaches some the proportional-plus-integral controller
limit, which may or may not be the limit of is written like this:
the manipulated variable. In digital con-
trollers this is a limit set in the menu for that
controller, or it may be set in the software. Kc 1+ 1
T:s
I
For electronic controllers it might be set
with a manual adjustment. For pneumatic
where Ti = integral time
controllers the nor1nal situation is that no
provision is made to avoid windup, but that Sometimes it is written this way:
extra instrument items can be installed to
combat the problem. 1js+1
Kc T:s
I
•
'
10
_._. ---
' ,, ·---' .
. '
·. . K .. - ''
,.• ,-. ,.
.
-,_f;,•_'
'' ' -..
. -·-
K +' c '
' '~
' '
.. . .c .....
. T:s
' ' '
.· ' 'contribution ·, ' of ' the·.·. integral action ·.. ... ·. .- '. ...,,1
-r:s .'
. -·-
--~--
:•· - ,-
' ;'
''
''
· . time is increased. ·
' ' ., -. _- .
'
'
.
'
' __ ,_
'
··. not have•a p~·in'the :i11'41f1.,.,'tinq term. ·.
. ' '
· · ·· · · b' • d · · ·
· Tuning. . rq/e$ ..artt " ..~~' ,, .g.n ·• . · the
Very shortly, · as you read other lite~
ture, you .will quickly recognize the · .· ''interacting,; tiltfljitgirn tor. ar Jea'St two ·
.· .reasons. First;. ··WheiJ···· tli!nin:g rules first
,_..
math: · ·.· ·
··were introduced by Ziegler and •· · 'S, · ·
.controllers were .built in the interacting .
1' " '' mode. . . Secbnd, .•.·.. when ' .· . frequency ..·
Kc 1+ T. ' ' '
I
·S •····.·response·analysi$ 'ilnd BOclej:Jlots came···
or · along it was much easier tc:J. understand ·
'
.· · What •was going on · if the interacting
-·---- algorithirf'was used. .·. · . . .• ··
'' . '
'
- . ·' .
- _, '
. '
-·-··-
'
,,
.'
:
'
'' - ' ,, '• '' ·:-:·,, .- _. '
. ; .· '
' ·,_ ' / -·. ' '
' . .
• •' - '· - '
' ' ,
f -
······-
' '.- ' .-·. . - -- ·-·--.
'_, ' '
'
' - :. '
' - ' '
---~-.--··
' '
'
·,;_.-, - ,. ' '
" - . --
'
'
. '
- . -' '
'
' ' '
'
- ·,,
__ - ' ,_ -'__.'
,,,_. ''
'
.'; ,' .,
-. '
' .- '
'
' '
'
- -., - ·-·-
' ' .
.
___ ,-_-
•
•
Fortunately, only a few ter1ns have evolved value and then decay back to some lower
over the years to refer to the function of de- steady state value. The amount of the steady
rivative action, and the scientific term de- state change is that due to the proportional
action only. You might ask why the
output is changing between its peak and
its final steady-state value, when the
error is not changing, and therefore there
""'y Derivative Component should be no output component due to
... the derivative action. This imperfect
=0 :JQ,
Cl) -
derivative action is a practical matter on
-=s
co - - - - - - - - - two counts. One is that it is physically
8 ..._ Proportional Component impossible to build a mathematically
-
I perfect derivativ,e function, . . and two is
, . • • . I - , ,
Figure 1.5. A proportional-plus-derivative controller will Figure 1.6 illustrates another way of
respond to a step change in error by adding to the conveying what the derivative function
proportional component that decays with time. The does. This time, instead of introducing a
longer the derivative the longer the decay time.
step change in error, a ramp change is
rivative seems to have held sway.
Rate ·(and Pre-Act, Taylor, starting
abOut 1940) have been used. I will use ... Proportlonal-plus-
derivative time and derivative action.
=...0 _
Cl) -
Q,
:J Oerivatlve Response "">,,i,..
- :J
It is mathematically the opposite of 50 . . •
0
integral action, but while we might I I. l\_Proportlonal-only
.• - Response
have an. integral-only controller, we
· I Derivative I
would never have a derivative-only . I I.-· Time ~I
controller (though we could have a
proportional-plus-derivative control-
I
ler, with no integral action). The
reason for this is that derivative action ... I
only knows that the error is changing. g I
w I
It doesn't know what the setpoint
actually is, so by itself it cannot
control to a setpoint. Time
Figure 1.6. A proportional-plus derivative controller will
Figure 1.5 shows the step response of respond to a ramp change in error by adding to the propor-
a proportional-plus-derivative control- tional-only response. The amount added will increase as the
ler. The output will peak at some derivative time is increased.
'
'
'
used. This is simply a change that continues derivative would have a transfer function
Of..
at a fixed rate, rather than all at once, as for "',
-~,,
.:)'.,,
""' -
,J-
:-"'
'
_;'
'
using enough but not too much, but if you denominator is the practical necessity. '
'
do not use enough, there is no benefit at all The.denominator is the transfer function ' '
'
and there could be some harrn. If you use of a lag, which will be discussed more in
just a little bit too much the troubles the next secion on niter time. The new
parameter, Kd , is known as the deriva-
increase a lot faster than the benefits. IF
tive gain. It determines the height of the
USED AT ALL, IT HAS TO BE SET peak in Figure 1. 5. If the derivative gain
INTELLIGENTLY. is 10, a typical figure, then the maximum
the derivative function can magnify any
rate of change is 10. '
1
'
'
The algebra for the derivative function It should be remembered that. usually '''
i'
gets more involved than what has been the derivative function on a digital ''
'
presented up until now. The ideal controller is set up to act only on the '
'
'
'
.,
''
•
With many digital control systems, ih.e so you do not see the waves generated by
menu for controller settings includes a set- the motion of the car.
i
ting for filter time. It is not normally in- . ~ .
'
I'
'
cluded in most published rules for tuning, The task of setting the filter time is one of
I
because when they were written there were using as much as you dare without degrad-
only analog (non-digital) controllers around. ing the performance of the loop. Too long a
The filter is a digital controller phenomenon filter time will· affect controller settings and
and helps compensate for the small also make the controller slower to respond
· to disturbances .. The use of it at all is
Actual Change in likely to start·.· a lively discussion
Controlled Variable
between those who grew up without its
availability; and · those who grew up
al Cll after its availa~ility.
=:s
..1: ·-..
0 Ill
A r. ..·.
.
o~ 0.63A
0
Figure 1. 7 shows what it does to the mea- It will get discussed more later. No-
surement; it slows it down a bit, or averages tice that it has the same form as the
it. The gas gauge in a car is heavily filtered, denominator in the proportional-plus-
derivative function.
'
·.' :-~·.;·-
'
..
- : '-
.
'
'
•
15
I want you to understand not only the me- combination at times. The open-loop
chanics of tuning (tuning by-the-numbers), method is a bit harder to use but yields more
but what you can expect from tuning. What
• •
fundamental TIM 11: and AMOUNT infor-
should you be looking for? There is a basic mation about the process. It is essential that
dilemma in explaining tuning rules, if part the persons involved feel confident in per-
of that explanation is committed to teaching forming the necessary procedures. With
what the tuning is doing. To understand either approach, open- or closed-loop, there
what the tuning can d<>, you have to under- is a section on what to do, and another sec-
stand the importance of lags in the process. tion on how to do it (procedures and
To understand the importance of the lags, techniques).The key is to upset the process
you have to understand what the tuning can enough to get the information you need,
do with lags present. It is like the control without getting into trouble.
loop itself where do we start? I have de-
cided to start with the tuning rules. If you Tuning settings . can also .be calculated. be-
are new to the subject this will almost surely fore or after
.
a loop
.
exists.. This writing. deals
require that you cycle back and forth in your only with tuning in the field (though tuning
reading between the tuning rules and the for level loops might be calculated at any
lags. time).
and how far the process is going to hurry or simply to leave the system as
respond to the controller output. Will you found it..
you be able to restore stability with
confidence if needed? You should make sure that changes are
communicated to all operating people.
0 You should know and agree on how If a log book is kept for this purpose,
• •
much change you will . allow in the use 1t.
process and in the controller. output.
Get all the infor111ation you can here.
•
The basic idea in the closed-loop method of change, most likely in the opposite direction
tuning is to get the loop to cycle without to be the safest. Once again, you are looking
getting into trouble, observe the nqtural pe- for any signs of a cycle in the process. Con-
riod and the ultimate gain at . that point, tinue this procedure of increasing the gain
and then back ()ff. Several preparation and testing for stability, until you see a sug-
items related to people and safety have gestion of a cycle. At that point you should
already been listed to be observed before change gain less than a factor of two, per-
any tuning activity should be perfor1ned. haps by only 50%, and continue the
Once those preparations have been procedure as before. As the loop becomes
completed the next step· is to tum the more oscillatory, make smaller changes in
integral and .derivative functions off, or in gain. If the controller design pern1its it, you
the case of . some controllers,
. as far toward may often safely change the gain while the
off as possible. This is to have the integral controlled variatye is moving. This saves
time as long as possible and the derivative making a new setpf>int change to disturb the
time as short as possible. process, since the process is already being
disturbed, and.that is·the purpose of the step
in the first place. '
I
,
• other is that the response will become more derstand trends and approximate
' oscillatory (unstable).
I relationships only. The effect of adding in-
tegral action is to reduce the error to zero
•
It is generally of little value to determine the (eventually), and to increase the tendency to
•
•
exact gain that will produce a steady cycle. cycle. The period of this cycle will be
The difference between that gain and one longer than that when only proportional
that produces a slightly decaying cycle is action was used.
small, and the period of cycling you will ob-
serve with a slightly decaying cycle is close
. '
j
20 Chapter 2, Tuning Rules and Procedures
l,
•
Figure 2.2. Integral action in a Pl controller eliminates the offset. have to know quite a bit about
Decreasing the integral time increases the tendency to cycle, and at tuning already, before you can
a longer period than for proportional-only. · use them effectively.
. .•
.
Find some way to 1·ecord the ultimate gain
If the perforn1ance with the selected settings
and the natural period, so that a person
is judged not acceptable it is not generally
following in your tuning footsteps will have
necessary, or even desirable, .to repeat ,the
that information. You might want to keep
full testing procedure. That is, it is not nec-
your own records as well. It would also be ·
essary to tum the integral ~d derivat~ve
appropriate ·to record something about the
functions off. The natural period, on which
operating conditions, such as the production
these settings are based, has already been
rate, setpoint and controller oUtput.
established. At this point it is well to
remember that the tuning rules are for
A final step before leaving the scene is to
typical loops. To build experience and
advise all who might be conce111ed, what the
therefore confidence in tuning, you may
settings were before the tuning effort, and
then alter the settings to modify the per-
what they are now.
for111ance in the direction desired, using the
concepts presented in figures 2.1 and 2.2 to
Chapter 2, Tuning Rules and Procedures 21
Kc = 1 I RL (aggressive) .
'
--·-.- • •
-· " - - '.-., - -
-- __ ,,,;;.
, ,·,,,_-
,
Figure 2.4 is provided to help discuss this This special situation that I am discussing is
situation. worth separating from the rest of the field
only if T63 I L is 3 or less (perhaps even 2
·or less). If it is, then the rules for tuning
'g GI
--
-.a
o ca
· are: ·
... ·-...
-a~ 0.9A A •
Kc= 1I3RL
CJ
Ti= T63
L •1---T. _ ____,...
These modifications to the rules have you
setting ·the gain lower than you· woulo
...GI- with the closed loop rules and the inte-
=2 :Ia. gral time much shorter than with either
·B
-- - -
c :I
oO
--.- the closed~ ·or open-loop rules.·· Thfs
CJ means, of course, that if you choose to
Time use these modifications, you . niust use
. .
Figure 2.4. The step response of a self-regulating proc- them together, you cannot use thetn to
ess can yield parameters helpful to guide tuning if the set just the· gain or just the integral time.
response time is short relative to the apparent dead time. Essentially you are going toward integral-'
only control as T63 IL become8:srnall, an
A few more parameters are defined. The option that . was .;ra:rely exercised before
step size is B and the final change in the digital control. The closed-loop rules will
process is A. A parameter called the 95% still give you stability and reasonably good
response time is as defined in the figure. perfor·mance, but they will .give you differ-
This response time does not have to be ent settings.
defined very precisely, as will become
apparent as you learn what is done with it. On rare occasions the slope, R, will con-
Ideally what we want is the 63% response tinue to increase. This :situation is often
time, but that may be hard to dete1·1nine at called open-loop unstable or conditionally
the time, when you don't know even what A stable or runaway. On other occasions the
is going to be. If you , are recording'. the ·. . . . r~poll$e, may go in the other direction first,
response, then it is possible to detetmine the ·..· called .. inve~se . response. The open-loop ·
63% response time from the recording, ~ut rules for tuning -do not apply in either of
still it doesn't have to be determined · these cases, and indeed, the closed-loop
precisely. procedure is subject to pitfalls then too.
•
.. ' -' -
·• ..
-
----·
' '
--.
,.,-,
--
-- •
'
'
- '
...
-~>-<··--_-_,_{i-'¢1
-- '·--~ ::,--
:'''"' : ,, _;; • •• ___ :<'' . ' ..
The open-loop approach to tunmg;r~quires You will have decided beforehand what size
' ' ' '
more care than the closed-loop approach, step to .introduce first, based on safety and
but yields more fundamental ·information. quality concerns. You may have chosen a
More care is required because the tuner may value considered very conservative. With
not feel comfortable controlling the process digital and other modem controllers there is
in manual, or may feel nervous about what likely to be no problem in making exactly
size step to . inject. The basic idea of an the change desired. With older equipment it
open-loop test is to learn certain 1'JM14: and is often difficult to make small changes
AMOUNT characteristics of the process by precisely. Do. your best. The important
.
thing
putting in a step change in controller output. is to not continue fussing with the.controller
output if the desired change is not achieved
For the open-loop test. to be .useful, the exactly. If you continue to adjust the output,
process must be running fairly smoothly. If trying to get exactly the desired change,
it isn't, then it may be too hard to differenti- then the upset is not a step, but is something
ate between what the step input caused, and else. If all this happtns fast relative to the
what might have been going to happen any- apparent dead time, then the ''sin'' is not so
way. So achieve reasonable stability of the serious. If it happens too slowly, relative to
process before putting the controller on the dead time, then the results may not be
manual. Most modem controllers have a validly interpreted as described.
bumpless procedure for· transferring from
automatic to manual: ·If the controller you To perforrn an open-loop test you must be
are working with does not, do the best you confident you can control the process in
can. It is very important that any bump to manual. It is very desirable to have some
the controller output at the time of transfer- idea of what R and L , and especially L, will
ring to manual not be superimposed on the be. Information given in Chapter•4 will help
planned step change in controller output. If you estimate L.
there is a bump from the transferring proce-
dure, or from your efforts to stabilize the
process in manual, let · the effects of it
settle out before trying a step test.
you will likely want to not return the output which could present problems. A good way .
to its original value, but rather double its to check for dead band is to .mue<two small
if<·.··.~- - .. ,_,--
size in the opposite direction, such as shown changes in the same directif>h,.iimd then re""
in Fig. 2.5. This format gives a balanced verse direction with the same two small
.
disturbance to the process. The intent is to changes. If the process does not repeat itself
end up where you started. This approach is for the same outputs, this is a good indica-
also useful when you don't know what size tion that significant dead band is present.
step to start with. You can start the step size Problems with dead band are discussed in
and duration at very conservative values and Chapter 8.
then increase both until you start to get the
results desired. Once the open-loop testing is done, calcu-
late and install the settings and return the
controller to automatic. At this time you
should ._ consider making small setpoint
changes and observing the response, much
as if you were perfonning a closed-loop
test, basically to confi1·1n that the results are
close to what you would expect. If the re-
sponse is judge<t too sluggish, increase the
gain. Try a factoE of 1.5 · or two. If it is
It is very desirable to perform the test more judged too oscillatory, decrease the gain.
than once, especially if there are a lot of Again try a factor of 1.5 or two. If it is oscil-
other variations going on, which casts some latory and the period is much ·1onger ·than
doubt on what you caused and what would the natural period, say· more than 30%
have happened anyway. Sometimes these longer,._ increase the integral time, possibly
other variations can be identified and by 50%. These suggested amounts are to get
stopped, by putting another controller in you started. The point is to not make small
manual. Other times the step size must be changes, like l 0% or 20%. You are looking
increased so the dominant effect observed is for a response which is essentially complete
due to the step change. If you are observing in three to five dead times, not 20 to 50.
the results on a monitor and have the ability
to make a hard copy, do it. This will facili- You should plan to monitor the perfo1·1nance
tate extracting the value for R. From these over a period of time to confinn. that there
collective results, estimate the best value for ··· are no problems.
the apparent dead time, L, and for the rate,
R. If these vary a lot, be conservative, use If the response is radically different from
the longest L and the largest R to calculate what I have shown, then it is likely there is
the controller settings. something abnorrnal about the loop, or that
the loop is being constantly hit by distur-
If the response is self-regulating, and if the bances. If making these tuning changes does
response time is short relative to the appar- not produce the expected results, and no
ent dead time, then use the modified rules disturbance that could produce the behavior
given earlier, essentially using a shorter in- is identified, then the judgment of a more
tegral time than you might when using the experienced person should be sought. This
general rules, and a somewhat smaller gain. of course asst1111es that the problem justifitfS
a deeper scrutiny and that other efforts to re-
If circumstances permit, it is a good idea to solve the observed behavior come up short.
make step changes of different sizes as well
as in different directions. If the response Observe the responsibility chores after tun-
curves are not in proportion to the step size, ing, related to staying available and letting
this is a good indicator of nonlinearities, people know what you have done.
!
'' '
The open-loop and closed-loop methods of The open-loop approach to tuning might be
testing for process dynamics will not give preferred as a precursor to the closed-loop
exactly the sa1ne controller settings, but approach when L is long, like minutes
they should not be very different. The open- rather than seconds. This could be the case
loop test is probably done more often when for composition and temperature loops, as
a monitor is available to display the com with press111·e, level and flow
controller output and the controlled variable loops. Indeed for pressure, level and flow
on an expanded scale,. than when it is not. systems, the L and R are likely to be so fast
This ability to exa•nine the open-loop that typically available recording equipment
response ''under a microscope'' has greatly might not be able to establish them
facilitated the use of the open-loop approach accurately, which could leave you using the
to tuning. The open-loop test is also very closed-loop approach to tuning.
useful for troubleshooting, as there is usu-
ally some idea what the step response
should look like, and if it doesn't, then there
are clues as to what the problem might be.
change of size B. The process gain, K , is · Ziegler was not considered one of them.
then said to be A I B. This ratio will ave About 1940 Taylor Instrument Co. (now
units of % process I % controller output. If ABB Kent-Taylor) introduced the first PID
the controller gain, Kc, is then set to less controller all in one case, and the company
than 1 I Kp (less than BI A), the loop has had to help its salesmen and customers learn
to be stable. Understanding this can be very how to tune it. Ziegler took on the task.
helpful at times, especially for those From talking with virtually everyone in the
situations for which it may be difficult to field worth talking with he learned that, in
establish a slope, R. for the open-loop his words, ''lags were bad and capacitance
response curve. This can happen when the was good.'' His use of the te11n
response is too fast to see on the equipment ''capacitance'' will need more interpretation
'I than will be presented right now, but a fuller
readily available. Sometimes A I B can be
calculated, in which case a value can be understanding of it should evolve as you
placed on gain that would be assured to read further. By ''lags'' he meant those fac-
give stability. tors that contribute to L, and by
''capacitance'' he meant those factors that
contributed to R.
'
'
.
.
From that understanding he conceived the reverse. If we have thus and so tuning, and
graphical open-loop method for dete1n1ining R and/or L changes, what would happen to •
controller settings. Many complex situations stability? If you are a novioe on control at
can be much better understood if time is this stage in your reading, I simply ask you
' taken to ask what would the open-loop to keep this open-loop concept in mind.
step response look like? What. would happen You will find it very useful later, especially
to R and L, and therefore, what would that in understanding the effects .of interactions
do to the desired settings? Would that make and non-linearities. These are the principal
the system more stable or less? It is rather conditions that cause a control loop to
like using the open-loop tuning rules in misbehave.
. .
. . .
There are two common times when going storage capacity. of the vessel to absorb
'
through the tuning procedure by-the-num- upsets. This is caped averaging level con-
bers is either not desirable or not necessary. trol. This is illustrated in Figure 2.6. The
These are level loops and flow loops. flow in can be quite ''wild'', but the capacity
•
Level is different from most loops because UL%= Allowed upper limit,% of scale
it rarely is a factor in quality control, as
flows, temperatures, pressures and compo- LL%= Allowed lower limit,% of scale
sitions are. Indeed, it is almost the opposite,
if you can conceive of an opposite. Often it SP% = Setpoint, % of scale.
is desirable not to have tight level control,
because this passes on upsets in one part of V =Volume between UL% and LL%, ft3
a process to another part. Rather it is desir-
able to have loose level control, using the Q = Maximum flow rate, ft3 I min
•
Then calculate the controller settings: On rare occasions level. falls in the quality-
affecting category• and ·should be controlled
Ket= 100% I (UL%.-SP%) · .· · . · tightly..One exa.mple of this is· a polymeri-
zation vessel in which the· shoreline builds
Ka= 100% I (SP% - LL%)· up a deposit of·. degraded polymer that
should. not be disturbed,
. lest it break off and
Kc= Smaller ofKc 1 and Kc2 contaminate the product You may think of
other examples. So again, you are back to
the tuning rules.
movement and level starting to respond is fall in the two to five second category. Since
short (say one to 10 seconds). If it is long flow . loops are. fast .· relative to most.
(and there is no hard and fast rule about the accompanying processes they don't. have to
one to 10 seconds) then· the considerations be tuned to within an inch of their lives, so
are more complex and you are advised to an integral .time of 0.1 .to 0.3 · minutes is
apply the open-loop concepts to the task of usually .adequate. These settings will get
i tuning. you in the ballpark..
. .
A case can be made for not using. integral at If you need to be more precise than the
all for level control. This is a more viabl.e above generalities yield, then continue with
concept now, with digital control, than· it the closed-loop approach to tuning, trying
was for. the days of predominantly analog small setpoint changes to test for stability.
control. For standardization on a plant, very • The open•.loop. response is usually too fast
few proportional only controllers were · to .captur~ accurately on typical convol
specified, with the result that PI controllers room. monitoring equipment, so the . open-
were used for level. At issue is. something loop approach is ··usually not usable;· Be .
that is hard to explain without getting more aware though of one factor when tuning
scientific. An integrating process, as level
0
flow loops. That is that the· open-~oop gain
usually is, coupled with a controller with is likely to be higher at high flows than at.
integral action, tends to look like a runaway low flows. The open-loop gain is how much
system .. This presents ·the. potential f?r the flow moves (in percent of scale), in
stability problems that I will try to explam response to a change in the controller output
later. A cycle may develop whose period is (in percent of scale). If this is higher at high
long relative to the natural period. · flows, as it often is, then tuning done at low
flows may become unstable at high flows.
I
Pause a moment to think about it. Processes could be worse than if a less customized ap-
come in a virtually limitless variety, while proach had been used.
controllers come with only two adjustments
to fit them. Oh yes, in truth there are three,
,froportional, Integral and Derivative (PID),
but integral and derivative are dete1·rnined
by the same process parameter, so having
set one, the setting for the other is also
deter111ined. Broadly conceived, the task of \
The reset time, the derivative time and the
tuning is concerned with fitting the time and
filter time are all keyed into the time pa-
amount parameters of the controller to the
rameter of the process. Unfortunately the
time and amount, parameters of the process.
controller gain is not uniquely tied to an
This is similar to fitting clothing (shoes are
amount parameter of the process; it is tied
a good example) to a human being (also
to both time and amount parameters of the
coming in a virtually limitless variety), by
process. Actually it is tied to a time-depend-
specifying only two parameters, like width
ent amount, but that is getting too compli-
and length. In both cases it turns out not all
cated at this stage. Just remember that the
that badly.
time settings of a controller are tied to a
time parameter in the process, so the time '
flow loop. . .
·.>.0.·:' -'·'''·'i/ti;~. _ -
' .·.; .-- -
- ' -
,·i ' •••.,,.c·
'i
''
'
(;
!(
'
i
'
''
31
Tuning rules are designed to give ''tight'' models for the process, and simple, well
control. This means they are set to give as defined disturbances. They present results
fast a response to setpoint changes or upsets defined to three significant figures. Real
as can be had without excessive cycling. Its processes and real disturbances are not often
like teaching everyone how to run the four- mathematically simple, and for disturbances
minute mile. That may be desirable at times, in particular; are often not mathematically
but frequently it is not necessary. Most of defined at all. Despite this, these studies are
the time it is not necessary. Sometimes tight very helpful for understanding generalities.
control is not even desirable, as has just So general relationships will be presented.
been stated with level
controls. Imagine a controller
on manual and that an
With the above upset occurs which
qualifiers I will now causes the controlled
concentrate on what variable to respond as
perfor1nance you can shown in Figure 3 .1.
reasonably expect to The figure also shows
get from a loop that what can be expected
is tightly tuned. Performance is judged by after the same upset if the controller is in
how closely the controlled variable is held automatic. Again the numerical values for
to the setpoint, both for setpoint changes the settings are to show trends only. Now
and for disturbances. People studying the comes a very important point. If tightly
problem mathematically have used a variety tuned, the controlled variable will deviate
of criteria. These methods take on scientific about as far as it would have without control
language, like integral of the error (IE), for a time a tad longer than Pn12, which is
integral of the absolute error (IAE), integral about equal to two apparent (or real) dead
of the error squared (ISE), integral of the times. An oversimplified reason for this is
absolute error multiplied by time (ITAE), that it takes one dead time for the controller
peak error and surely others. These studies to know about the upset and then another
all assume certain mathematically simple dead time for it to do anything about it.
•
1. T1 = ''Off'' '
2. T1 = 100
3. T1 = 50
4. T1 = 25
All curves with K0 = 2.5 5. T1 = 13
Figure 3.2. Integral action restores the controlled variable to the setpoint.
Reducing the integral time increases the tendency to cycle, and at a longer
period than for proportional-only control.
Chapter 3, Tuning Objectives and 33
Expected Loop Performance
I had hoped to get your attention earlier by scale, be able to reach the desired gain for
emphasizing the importance of the natural stability.
period. What you may not have;: noticed was
the absence of any emphasis on the specific
ultimate gain it took to cause the loop to
cycle. This was deliberate. Whether the loop
cycled with a gain of 0.1 or 10 tells you
very little about the expected perfor1nance.
In the example just given the upset was not able to do a pretty good job of
defined, either numerically or in kind. It was compensating for it. Of course you may not
implied that the upset came and stayed. The have time to wash, but that is a separate
concepts presented were important. There is problem!
· another concept which is also important, but
it must be presented in a different way.
,,.--~--___..,
- /, .....
--; -
__
-... ..ec
•
I : \
..c 0
·O
/
0 0
0 ..
•• --r 1
·-..ta
0
.c :I
..·-;: .c..0 I
\
a:
-0 ·-;: Natural I
1 ' 1
Cl
_,0
G>
'C
..- ..
G>
Period-,,
I I ' 1
:I 'C
:I I '
0.
E
-0.
I '
ct E
ct
I
I
' ~Higher Gain
Figure 3.3. Control action can either help, hurt or do nothing to. reduce the
effect of a disturbance, depending on whether the disturbance is slower, at or
faster than the natural period. ··
The part to remember is that if the natural period, the untrained per~on is likely
disturbance is periodic, the control system to conclude that the controller is not doing
can either be helpless, aggravating, or its job. You need to recognize this situation,
helpful, depending on the relationship of the infor1n
.
the appropriate. persons . that more
disturbance period to the natural period. efforts on tuning will not do the job, and
Look again at the two equations for tuning: Combine these two equations and you have
•
the most important simple concept ID
Ti= 1.2 Pn for the closed-loop testing control:
g~
Pn = 2L
great insight it provides. The symbol : : , (,)
meaning approximate, is used because the L Time
tuning rules are approximate. Actually, the a. Pure Dead Time Only
use of the tuning rules to ''prove'' this
relationship is not valid. The tuning rules
are based on more scientific principles and
"i
-- Cl>
-- .c Pn = 4L
the proof is in those scientific principles. -..g~·-..
0 IV
(,)
Time
The above relationship of Pn to L is an
approximation. It is possible to be more Pure Dead Time Plus Integrator
precise. In Figure 3.4 are shown two
extremes and something in between. If the "i
-- Cl>
-.o
step response of the process is as shown,
then the relationship given to the right
-..g~·-..
0 IV
2L < Pn < 4L
(,)
applies. It is possible to confirrn the first
Time
example (Figure 3.4a), by logically
drawing out what happens as a function of c. Mixture
time around the loop when it is cycling.
Figure 3.4. The natural period varies between 2L
Refer to appendix page A-2. This one is and 4L, depending on the other lags in the
easy, and the equation Pn=2L is exact. system.
'
:;
a.
= I
0
I
I
I Now you may feel as if you've learned
I something, and you4tave, but as soon as you
=a.c try to apply this knC:wtedge you will have
- more questions. A few processes have a true
dead time, but most · do not. So what
Time contributes to the apparent dead time? Well,
Figure 3.5. When a loop is cyding the output of first the answer and then a long explanation
an integrator lags the input by 90°. of terms and examples. The apparent dead
time is any pure dead time plus all the little
the integrator output is moving at its fastest lags in a control loop. The paradox is that
rate when the input is at its maximum. The the largest lag in a control loop, sometimes
result is that the output of the integrator lags an integrating element, has little or nothing
the input by 1/4 of the period of the sine to do with the natural period.
wave, or 90° as shown. It will always lag
the input by 90°. The amplitude of the
output wave will not match the amplitude of
the input wave, as ·it did .for a pure dead
time, but the phase lag will always be 90°.
This second example, of dead time plus
integrator, is shown on appendix page A-3.
It is admittedly harder to follow the logic of
the signal around the loop than it was for the
pure dead time case, but try it. The
This is a difficult concept to accept. To the
relationship P = 4L is exact.
untrained it seems almost contradictory. It is
reasonable to think that the speed of a loop
The last example, Figure 3.4c, is more like
will depend on the lags, but it is
real life, but no illustration comparable to
unreasonable to think that the largest lag
pages A-2 and A-3 is provided. I appeal to
has nothing to do with speed (of the loop).
your sense of reasonableness that Pn will
Some may be asking about now, what a lag
fall between 2L and 4L. Place permanently
is. Let's just leave the above paradox, go off
=
in your memory that Pn 4L and you will
to understand some lags, and then come
cover a preponderant majority of industrial
back to try to tie everything together.
38 Chapter 3, Tuning Objectives and
Expected Loop Performance
' '
·-
' ' . .. -
.,
idealized, has no phase lag. A pure dead
time has a phase lag of.·
When the loop is cycling, every ppint in
the loop is cycling.· When· the cycle gets
35
back to any point in the loop it started Phase lag= oL
p
from it matches the cycle there, so the
cycle sustains itself. ff it did not match, in
Where: L = dead time
amplitude and phase, it would not keep
P =period
cycling in a steady manner. ·Either the
· cycle would grow or decay.
. . This should be easy to understand.
When the dead time and the period are ·
There are methods for. determining
the same, the phase Jag is 3600. So
whether the cycle will get worse .or
what is P when the phase lag is 1800?
better, but they are too complex to tackle
You've got it!
here. The phase lag around the loop is
·. 3600 when it is cycling, or the phase is
P= 360L = 360L = 2 L
minus 3600. Half of this comes from the
contro.ller itself. NotiCe in the signal-flow Phase lag 180
diagram for the controller (page A-2) that • •
the ·controlled variable is subtracted All you have to do is adjust the controller
from the setpoint. · gain to make the amplitude right, and a
cycle at P = 2 L will result.
This minus sign has to .
be' else the
c~ntroller would ·act .· in the wrong Now lets take the example on page A-3.
direction. The minus sign inverts the The integrator has a phase lag of 900 all
sine wave, which ends up being a phase the time, at any period. So the dead time
lag of 1800, when the loop is cycling. has to contribute only 900. What is P
This leaves 1800 to be contributed by when the phase lag is 900?
the lags.
P= 360L = 360L = 4 L
When the loop is cycling the contribution Phase lag . 90
of each element to the 1800 lag is a
specific value, as well as another Marvelous! Was that hard? ·
parameter, the amplitude ratio (ratio of
•
39
•
'
The word lag is general. It means any are close enough. There are other building-
relationship in which some result happens block lags which the more experienced
after some cause. There is an associated practitioner may use, but the need for them
word, lead, which in some respects is the is rare and they will not be discussed here.
opposite (though no result can happen
before some cause). It will be discussed
more later in the special section on
derivative action. You cannot really In the typical control loop, lags act in series,
understand control without understanding the output of one being the input to another.
lags, but there are different depths of For instance, the lags around a simple
understanding, and it is possible to get a temperature loop might be:
useful amount of understanding with a
modest amount of training. No proofs will + The output of a controller is the input to
be given. Most and possibly all of the facts a pneumatic transmission lag.
used are consistent with reasoning powers.
+ . The output of this transmission lag is
the input to a valve lag.
FIRST A FEW POIN'I'S. + The output of the valve lag is the input
to a process heat lag.
Figure 4.1. For the overall lag, it doesn't matter whether the component lags are in their proper
sequence or scrambled.
In this case six lags have been enumerated, noninteracting case the level in the second
each in itself quite complex if you were to tank does not affect the flow out of the first,
study it closely. Their net series effect can in the interacting case it does. The point
be represented by two or three building- made about being able to scramble the order
block lags. That was the last point. This of the lags applies only to noninteracting
point is that, · as · long as you are not lags, interacting lags must be treated as a
interested in some inter111ediate · point, it single dynamic element.
doesn't matter what order the lags are in or
what order the building blocks are in. A This is more of an academic point than a
signal-flow diagram is · shown in Figure real one at this stage in the development,
4.la. The order of the lags may be
scrambled, as shown in figure 4.1 b, which
makes no physical sense, but the lag from
beginning to end will be the same,
scrambled or not scrambled.
a. Interacting lags
II\\
·.·because
- .--- it i~~;~~l·ik·e·it';;;~o·u
:'7'_:_~JJ.':.~--"-~~-~-,;,~---~~- wo··u······1·d···,.,,.nt +o···:·•·····
-~----·· ,.,..,.., ·.-::~----.- __-:_~_,!·-·_,-:.· _ ··•···· ······· · __ ···.:___-___·····
___ ·;-~:. -.-''----~----- ·_·::·;;_:_;__ ·--.-:.:-.'-"'· ·· .. ····••··
,: ____ -_·::·-_._,_--·. ",,
· __···•-..·_·· __ · ·_--. . ·-. ·· ' · .····
- " . · ·----
• scramble the .6'·· .~;;if::'. ··:e tanks·ht this' case; ' ••.• .• . f/il\lles 4.3 !tti 4.$ 'are·. provkf8;'t0 .:ru11strate . · · ·. ,· ;
1t ·is. still an···· · · • ····.• • ···;J;t ·. . •. ·. ~ hf;'\1Vevel" . ···tiJ· •;·•·· ·.· ·. .~e step•tesl>Onse of eaclt' of th'e~ ·:~$ ·~t ·.· ·'
0 1
The following names are given···to the three In tl}e mathll\lg~~a s.•ctions l will introduce
types of lags that may be combined to their Laplape.~ans(04:1os an,d giv,e you afew
represent most physical syste1ns~ fonnulAs fQt·.Pli~ •d ·'1ltplitllt:le 1 ratios.· I
will use tile~ to illustrate a few . p0ints.
. .
' - ' .f"'. __--.·_··-:': -
•' -i:
--. ·- -•·
, __ .., ......
- . - . '
I
•
'
.
- -
•
'
... -- ' .
. . '
•-'- - ' ,_,_ -
-.----->- ,,_,-__ ;
,, ,' -
__ .,_-,.,-.,,
.. ' . I ', ·.·, ,
.--·
' "; ,• ,•
,•)
•.;
- • •
' -
I,
- ....
' ". •
·•
' •
'
"'·-···-
- :·-·-·.;;:·_'· :_·- ·-·--·· -·----~:-~·-;_'.;,._ :'.:~.·::·
•
'
;'',·'·c'<'''
- . . ,-
,' ,,,, _- ,· ·. ·- -
..
', ' - .c . ,•:-'.
: .. __. . . ·,·,_;;,-c, ' '
,- -
..- .-----·-;· ·.{
- - - - .
-,.. ·' -
- : ,'';i,. ·' - -
-. I • . -
.
' .
.
.. . .
···:··.-.-··-
_" : . · ,-~-
. ... '- -
.
.
. .
' .
-· - . ' ..
. .
'·- . __
•
"' .
' -- ..
. .- ' .
,.,. ''•'
''
. ..'.. .
'
·- - . ..-
••
. ' '
.-.' - '
" ,·_..
. " ..
'
.
-- . .- ' -'
. .
..
' - -, •
' .
'
'
The dead time step response is shown ~Time of Input Step Change
This next building-block lag goes by many Its slope at any point is such that it would
names. It has been called a time constant reach the final value in a time called the
lag, an RC lag, an exponential lag, a sill1ple time constant. It actually goes 63% of the
lag, a single linear lag, and surely many way to the final value in one time constant,
others. I will call it a first order lag. This is 95% in three time constants, 99% in five
a somewhat arbitrary choice, based on what time constants. This holds at any point on
seems to be gaining in popular usage. The the curve. If it had gone 23% of the way, it
differential equation that describes it is first would go 63% of the rest of the way in one
order. This is still a terrn that will mean time constant. Much more could be said
nothing to you unless you are schooled in about this one's characteristics. It results
differential equations. Its step response is from a resistance to energy or material flow
shown in Figure 4.5. and a storage for that energy or material. An
example would be the temperature in a
(
___ /
' ' '
.'
a change in wall tempera- '.·:·.,-·
..
-.......,·
./' '
'·
' '
.
' '
'
' '
·.~· -
ture. -
:::I
/ T = Time Constant
-0
Q.
/ t 100°/o Laplace Transform:
:I
~ I s3o/o
1
The first order lag is a
Ts+ 1
cross between the pure -T--1
dead time and the Time
integrator. The way it
responds to a sinusoidal Figure 4.5. A first order lag will respond exponentially to a step input
input depends on the change.
period of the oscillation
relative to its time constant. If the period The phase Jag is:
is short relative to the time constant, then
the phase lag and the amplitude ratio will 1
Phase Jag= tan- (ro T) = tan-1
6.28T
p
times the apparent dead time, so a further So now I hope you have some insight
approximation exists in predicting loop leading to acceptance of the par1,ldox, stated
perfo1·1nance. That is the penalty paid previously: it is the small I£gs~ and dead
without getting more complicated. time, which detennine L and, therefore the
natural period. The largest lag or the
A useful analogy for understanding the integrator have little or no bearing on
way lags combine is the way vehicles start establishing the natural period.
out after a traffic light turns from red .to
green. Each vehicle responds after the one
preceeding it, similar to the way elements A paradox!
in a control loop respond to the one
preceeding it in the loop. Some drivers
respond slowly, some more quickly. If you
are at the end of a long line it doesn't
matter whether the slowest driver was first There is not much more I want to say
in line, in the middle somewhere or just in about the math I algebra at this point
front of you, the rate at which you can except to point out the similarity of the
accelerate is the same. first order lag to the dead time when the
period is long relative to the time
Since you now know how to estimate the . constant or dead time. The following
table shows these similarities:
apparent dead time from the individual lags
in the system, you also know how to
estimate the natural period. Actually the Phase Jag, Amplitude ratio
step response does not even need to be '
degrees
drawn; you can calculate the apparent dead TIP First Dead First Dead
time directly from the estimated lags. or order time order time
LIP Jag lag
"' "'
""""""
''
"""'
··-·'. ·--' "'
''""" """""""' '''"' -·- -------·-
···--·-·- -- -- - .......... , ... ' " ' ''
0.01 3.59 3.6 0.998 1.0
0.02 7.16 7.2 0.992 1.0
0.05 17.4 18.0 0.954 1.0
' · ·-· · ··· . · ·· · · -- .· a··n··d·· 0.10 32.1 36.0 0.846 1.0
::'>:: ::::,,:::._' ' ' ·:.::_ : ,', ' ,::_:·_ ,.
""' ' - - '
- -_ -- '
'
' '. .·. . ..,,.t.'
- ---- ' ... h·····e····'·:r·······e'····:110····,:r·······e····
', : : .,:;: ::
·;;;; '"
:·
.
: '
.. .... " •···-·- '
: .:.
''' "'
·:.. '
. :· ., ' ;_·
'
- -' - ;;_ .::- ;· ___
- -_____-·-- ··•; ·
----------- ' ·· · ··,,,,,,,,,.
:;;,;,,,::,:,,:::::::::::::::::··:.:::···,,· ... ::::::<::::::
-- -:. "::::"::."::·::::::::.
·;;_;··--
"' """' """"
- -- --
---- . --- .
" '
"'
"' " " "
-" '
' "
-···-- -··
"' "' "" "
""""""""" """""'"""
-· '"
- - - -- -- -- ---
---- --· - - -- -·-- - ---- ---- - -- -- - '" '" """"'
""""""'' ........
....." ' ',_
"'
' ' ." -
- -- .
__ ,,,,,-
, ---------
... -
- .------
--·------ --·
Assume a loop has a dead time element
----------
- -- -·---·-- -
'
-- - •
' --:.;:., - ,_
--," -.•-,;:.--.. _ --
__
I have stated earlier that what the controller confusing at best, I will say a little more
gain turns out to be is largely irrelevant. about it in the math/algebra paragraphs at
There is another gain, which I call the the end of this section.
process gain. It bears directly on what the
controller gain will be, but since I have said A person not trained to separate time from
the latter is not very important, why discuss amount is likely to use what I would
something that . conttihl!tes to it. It is consider the wrong word to describe
important, for the completeness of your something. He or she might say something
understanding, to realize. that, processes not responds faster when I would say it
only have lags (of various types), but also responds more, or has more gain. Figure 4.8
gains. It is an · important • part of your is an example of a first order lag. In each
vocabulary, the language of control. case the time constant. is the same, but one
"
.'
;!•"
0
,,,- .
·:---,,',
-- - -.
' '
- '-.;-,_:[.,; ""'
Until now I have bfilff"~~ .· Kp
- ':4.": ·, first order lag
'I". .
Laplace transform as only · •· · . .· .. 1S+ 1 ·
dependent part. To be . complete they ·
should heve been · written with the 'Wittlotit
1
.. . .~..... . Kp, tren""'-..
the·· 'ilHJITI. fu nctJons
period-inde ·nt .part i ·· ..· .· . ; · · This are ·. dimensionless. . T and L have
amounts to simply including Kp in the dimensions of time, and s has
transfer function. For the three Jags dimensions of reciprocal time, so their
discussed then, their transfer functions Pf!Jduct. is dimensionl~ss. Kp will have
become: d1mens1ons. For the integrator this is
-·.
-
,-... _;;·.
.
:··
.. ,. ----:-•,
'
-
-;•· important, because at first it looks like
Kp e -rs, .dead·-: time · one number over another number.
'._ - .. Keeping the two parameters helps keep
the dimensions straight.
·. .
.
. . .
. .. . .
- - -- '
.
•
.
'
51
- - ) '-
' '
' ----'-.''"''· '
-:-:·
'
---
' ' ,· - •" .- ,.
-- ' ''' -- - ' '' •'
'
'
·- - -
' '' -
Probably all pure dead time lags involve a am not aware of any quick estimating
distance and a. velocity. Generally both technique to determine how much lag this
distance and velocity are well know, so the might add; it · would depend on the
dead time is well known. The velocity may parameters of the system. Generally it is
change with process ;conditions, but a new small relative to typical temperature
dead 1ime can be calculated. One example measurement lags (to be discussed later) ...
would be coating weight on a moving film.
Another would be the actual weight after Sample lines for analyzers, such as for IR or
casting or forming the sheet· in the first chromatography, introduce a· dead time. If
place, such as in polyester sheet.material or the flow is turbulent, · then there is . little
in paper making. These dead times tend'tQ ' longitudinal mixing and the dead time of
run to minutes, because of the distances and composition• to· the. 'analyzer .is virtually
velocities involved. ·· pure. If the' flow is laminar, such as in a '
' temperature measurement point creates a has a lag beyond that as the material along
''
r dead time. With a typical design velocity the edge of the pipe comes along, but this is
r'•
••
''
'' ''
'
•
•
not pure dead time. Generally, in the connection with Figure 3.3, it may tum out
viscosity case, you won't be far wrong by that the lag is too long relative to the upsets
assuming pure dead time, calculated from expected. The point I want to make is that
distance and average velocity. The the dead time, per se, is no problem. It is the
consequences of this simplification depend speed and severity of the upsets relative to
on what other lags are in the loop. Typically the dead time that are important.
it would be a conservative assumption.
.
-
:,
- --- ,,,. -
.,,•''
'
40 - - - - - - - - -
30 - - - - - - - - - -
20 - - - - - - - - - -
/
~-· }!'-·--
.
-~- - ~,,,..,,, ...
, .• '?;·__ ._ -.
"'''''"''"'·'·'
-- - -~--"·'-- -
--
--
- --
'
--
--
,., --,·-l -~-
/
Lags of transmission lines are· ,. ·e:,· ·
' - .. - . - .
•
•.
•
of a l09p, ' ' ional··that is to the expected .·as such. lt woul4·.~ exactly first: erder if
degradation of natural period. It deceives flow through the exit restriction were
the observer intQ .• . · ,, .· . f.h~t .performance
··-· - ., ...." .• ,1'.;.f .,., :·-·''"' ' - '
laminar
- -- ''··
(flow lil}ear with
---,,
head).
; - _- ',•: --·-
are always . larger tha~ ,:~ . . ·:'1,v~~ lt is the11. apparent that the.actual volume. in
SometiJiles this is not important; sometimes tj\e ~~··i$ irre~vailt~. -. What is.relevant is the
·- ·...___ ,-,_._..;·:-:~·"·--_.-,--,~-~---··: ·-.~- -· .··'"' ·.-.. ._ -·
it is'I, O~asionally
-- '
profound.
-
~provements - -
change i11 \ro~J,Ime fqr;;a .. . . · Pi flow.
in . control perf()r·n1ance .·can ,be . achieved This approach~pplie~ even if.the level is on
through a faster measurement. ·•.
' '
proportional .. only .control
--~ '
and·.·.
-
is much~ --- ' '
derivation. ,_,-,
.. -
-
''.-
' '
-'
'
---
~ '
-
' '
'
Assume that each tank is · holding 200 In the system shown in figure 5.3 the
gallons and that ' the otittlow ' depends composition in the tank will follow the feed
linearly on head (is not pumped out), It is composition with a first order lag calculated
from the hold-up volume and the flow '' '
'
'
tempting to say that the ·lag is 200 ' '' place, all bets are off for any simple answer.
gallons/100 gpm or two minutes.. It is also The differential equations represent the only
wrong. This concept applies to composition road I know of to the solution.
but not to volumetric flow. If the flow in is
increased to say 120 gpm, the level will rise
until eventually the flow out is the same.
While the lag can be derived from the
differential equations, most persons would LC
shy away from that approach. Happily it can
be deterrnined a simpler way. Calculate
what the new level would be, based on the
new flow. From this, calculate what the
change in inventory is. Assume this
Figure 5.3. The lag of a stirred tank to
calculation yields 10 gallons. The lag is then
composition changes is its volume
10 gallons/ 20 gpm or 0.5 minutes. It would divided by its throughput.
be close enough to a first order lag to treat it
•
This is a rather broad-brush category meant While there are exceptions, control 'loops on
to include everything but measurement lags. the same type of variable tend to have the
Process lags, such as heat exchangers, are same natural periods. Flow loops will have
not only complex, but the variety of a natural period of · 1 or 2 seconds if
equipment is broad. These factors combine transmission lags are short, going up to 5 to
to discourage trying to condense the field to 10 seconds for long pneumatic lines. Flow
typical numbers. Lags get longer as the should be back to no1·mal, after an upset,
surface-to-volume (mass) ratio decreases. within a very short time, say 5 to 10 seconds
They cover the gamut from a few seconds to for the faster loops. Pressure loops fall in
hours. the same category. They will be fast. The
,. '
57
.,
'.' . ; __,_
-
.'' - ' • > ''
- ·- .
- ''
' .
Supply
·.·Reactor ·
If you . have never been exposed to a Brine
cascade control system it is confusing at Exhaust
•' ;.
- ' ' 0
'
-.; - :;•.
Setpoints
__
. .;,.-·--
,,.-,.·' '
Process2 Process1
- +
Controller 1 ·(Jacket) (Reactor)
- -
Inner Loop
Outer Loop
Figure 6.2. The signal-flow diagram of a cascade control system shows the inner and outer loops,
•
then the brine temperature controller will and to some extent on the ameunt of metal
know about it sooner and take corrective closely linked ther111ally to it. This can be
action. Without the cascade system, no relatively long. Now consider the cascade
knowledge of the upset due to brine header system. The reactor temperature controller
pressure would exist until· it'had affected.the "· · oµtput wants a temperature, oot a valve
reactor temperature. · This '~· inu··
· 1·a·.. · • b. ··. .... ·.. position. The secondary.. contrJ~ler gc;res
. . .. . .~ . e .
substantially later, so the .disturbance would ·..... after that request by doing all it. can to get
have been present for a much longer time. the temperature requested. The result is that
the jacket temperature changes much more
rapidty than it did to a simple valve position
2. It will effectively lirtearize the secondary change. This result, of course,· reduces the.
variable to a change in setpoint from the lag the primary controller thinks is in the
primary con:troller, and linearity is generally loop and so the . natural period af the
desired for optimum performance. ·• primary .. loop is ·reduced. Refer again to
·-." '
Figure· 2.2. The overall lag to . a setpoint
In the exa1nple, a 1% change in the primary change is less as the controller is tightly
controller output will result in a defined · tuned, than it is when it is not.
change in jacket temperature (because the ·
brine temperature controller will make it
~o). W.ithout cascade control, a 1% change · .
4. . Some··· instrumentation
- '
tricks can be - '
i~ the primary controller output, . going .· · ·. •. played, .such as incorporating limits in the .
directly to the brine valve, will re.suit .in· an •. . · · secondary ·set point.
' ' -
•'·' ,•-'
gland friction. ,• -- -
..
3. It can, under mfllly and perqaps mo~t
---- . . --
circumstances, shorten the natural period 0£ There is · an important . dynamic
the primary loop. If the secondary loop has . . consideration when contemplating whether
a lag in it that would also be in the primary to use cascade control or not. The natural
variable loop without cascade, and if this · period of the secondary · loop should be
lag would be of consequence in establishing significantly shorter than the natural period
the apparent dead time . in the ·primary of the primary loop would be without
variable loop, then cascade control will be cascade. There is no hard and fast rule on
faster than no-cascade control.
- -- -------- - ·------ - -·-·-- -- - --
the desired difference. A factor of ten is controller, like another lag, or combination
certainly acceptable. A factor of three of lags. The amount of that lag depends on
becomes questionable and a factor of two tuning. In this sense, the tuning of the
becomes very questiortable;,.:for the latter secondary loop interacts with the tuning of
situation to be attractive, c '' . tiQD.s the primary loop, but not vice versa.
other than dynamic would have to dominate
the decision.
Always tune the secondary loop first. It then
becomes a lag in the primary loop. In
cascade control systems, we nearly always
tune the secondary for tight control. It
would rarely make anY ·sense to have a
then many of the benefits of cascade control simplify maintenance and replacement,
would not accrue. Then tune the primary these considerations often overrode
loop as you would any other loop. The thing performance considerations. Now, with
to remember is that the . amount of. lag seen
. digital controllers having the capability of
by the primary loop depends on the tuning being either PI or simply P only, the pros
·of the secondary loop. Refer again to Figure and cons of using integral action for the
2.2, to remind yourself that the response to a secondary controller should be reviewed.
setpoint change simply looks, to the primary
61
•
•
•
_,,, i·. ' --· •
-., ~:.-_-:··,.----- , .
'.• .. ,o"
-.. -·,-,- '--~'.- -- ' :
..
Derivative action is fairly tricky to use The following are the most important
successfully. It is very helpful to have at application points about the use of
least some appreciation of how frequency derivative action:
response analysis applies to closed loop
systems, but most people for whom this 0 It will decrease the natural period of
booklet is intended will not have that. So almost any loop and, therefore, improve
its performance. The improvement in
the natural period normally is modest,
being in the 10 to 20% range, reaching
perhaps 30% in rare instances.
No1·mally better tuning of a PI controller
will yield far more improvement than
that. The effect of derivative can be
what can be said to be helpful, without visualized in the step response shown in
getting in too deeply? figure 7 .1. It is seen that if the normal
step response has somce curvature before
The rule for setting derivative action is to the . main response,
. .
the derivative can
set it equal to Pn/8 (L/2), though some accentuate this and in effect reduce
.
the
recommend the divisor to be six. The point apparent dead time. This in turn reduces
is that there is not much room for error.
Use too little derivative and it is hardly Proportional plus derivative
worth the bother. Use too much and you. applied to normal response /
can create stability problems. The ratio of 3l . ~
too much to too little is relatively small, c /
like perhaps 3 to 1, or possibly even 2 to g_ / · Normal response
xi / without derivative
1. It is not unusual with a pneumatic cc /
,,,.,,.,.
controller to have the calibration of the / ,
derivative adjustment in error by a factor
Time
of two, so that makes it particularly
difficult to use properly. It can make
heroes out of the knowledgeable and bums
L, - -
out of everyone else. That is why it is
seldom specified and often is turned off Figure 7.1. The effect of derivative action may be
visualized as reducing the effective dead time.
even if specified.
•
the natural period which is not now its amplified no more than that gain (the typi-
''natural'' period. cally 6 or 10). The step response to the typi-
cal implementation of proportional plus
0 Derivative action can make a runaway derivative action is fairly easy to visualize
process stable. Once in a great while and is shown in figure 7.2.
there will be a process that cannot be
controlled in manual. It can be likened . ·. It is a spike whose gain is K, coupled with a
'
to trying to balance a long stick on your _. ·. time constant decay of gain K-1, and time
finger. If you hold your finger ·still·· constant T.
(open the loop of eyes, brain and
muscles) the stick will fall. The system ,__T I
I
is called unstable, not bOpa~e it cyclCs, I
but because it will not stay stable. Ifthe - - -1- -- --t - - ·-
loop is closed using the brain as a I 0.63 (K-1)
controller between eye and finger, the 3lc K
I K-1
stick can be held upright. Derivative 8.
a:=
action is not essential to stabilize
runaway process, but it can help, 1
sometimes significantly . +-
..
Time
0 ·For a batch process it can help ''tum the
comer'' without overshoot (undershoot), Figure 7.2. The step response of a proportional-plus-
especially if the integral and derivative derivative controller will spike up at first and then
decay back to a steady state value.
comers (these corners are frequency
response lingo) are exchanged. This will
require that the no1·inal controller gain A mathematically · pure derivative function
be reduced by the separation of the would respond to a pure step. as .a pulse,
corners. There is more to this than I infinitely ·high, infinitely short, · but
want to get into here. enclosing a.n area equal to or proportional to
the derivative time. That is a rather abstract
0 For a batch process, I used it once to concept. Actual implementation of the
compensate for a long pneumatic derivative function can be visualized as
transmission lag. A vessel was filling or trying to get this area under the response
emptying, so the signal to ·close the curve without going to infinite amplitude.
valve was based on level and its The lower the amplitude, the slower must be
derivative. the decay to get the same area. If the
•
'
•
problem· in a later. section, called 5,ainpling Here · coifHiS a bit <#. algebra. · The
Frequency
'
and.
. .-
'
..
.
_-
, LOap ·.
·.
· Perrorm~ce, in-
response·· shown ..Jn figure ·7.2 may be
considered 8$ made upc#two parts, first
' ' .,.~----- '
times per second, you should question the back. The algebra of this is:
use of derivative action· for processes having
a natural period shorter than about a minute.
' '
' ' '
'
''
65
'
'
•
•
Interactions and nonlinearities make control particularly helpful to think about what is
loops interesting. Loop behavior in their happening to the open-loop step response in
presence does not follow the nice clean terms of R and L, and then interpret that in
patterns I have described. You have to know terms of what the interaction or nonlinearity
the nice clean patterns so you know what would mean to stability. I encourage you to
you could expect if it were not for the inter- do the same when you are faced with
actions and nonlinearities. I have found it something that does not quite fit the norm.
Consider the system shown in figure 8.1. of the other. The interaction in this example
is particularly bad because the natural
periods , of the two loops are essentially
PC FC
identical. One simple helpful approach for
understanding interaction problems is to
reason what happens to the open loop step
response of the other loop when one loop is
tuned tightly. Then apply the concepts of
Figure 8.1. Tuning of interacting loops can be difficult. the Ziegler and Nichols open-loop tuning
rules to this open loop step response to see
A movement of the pressure control valve whether the desired gain or integral might
have changed, based on Rand L (and RL) in
affects flow. A movement of the flow
Figure 2.3.
control valve affects pressure. The two
control systems are said to interact. The
tuning of one will affect the open loop There are several approaches to combating
interaction problems, and the following list
response of the other and, hence, the tuning
is not all-inclusive.
•
0 Decide whether one of the loops is su- there would be any problems with the two
perfluous. This does happen. For in- loops talking to each other. So once again it
stance in the example given, is the pres- is essential in solving cohtrol problems to
sure loop really needed? know what the natural ~riod is, what
detennines it and what can be done to
0 Rearrange what valve is manipulated by change it.
what variable. This could require more
insight than the novice might have. • To repeat what has already been said, but
with a different emphasis interacting
0 Decide which variable is more systems can be tricky to tune. It is possible
important and tune that loop to be tight. to arrive at one set of tuning parameters for
In the example given the flow loop .one loop if the other loop is on manual, and
might be made the tight one·.· .· . a different set if the other loop is on auto-
matic and tightly tuned. It is possible for
0 If a computer-based control system is one loop to become unstable if the other
used, it might be practical to program or loop is switched from automatic to manual,
configure decoupling tem1s to reduce or vice versa. This. section on interactions is
"'
the cross• talk. Again, this may be too more to alert you to the potential problems
complicated for the novice to undertake. · than to equip you to solve them. Basically,
unless the problem deserves a more elegant
Almost implicit. when there is a problem approach, I recommend tuning for the worst
with interaction is that the natural periods of case (those conditions that are most likely to
the two loops are similar. If one loop has a produce cycling), and accepting the per-
natural period of one second and the other fo1n1ance at other conditions.
of one minute, it is highly unlikely that
Until now, I have assumed a linear system. Nonlinearities come in 11lany fortns, Some
Mathematically this ·means that I can de- can · be troublesome. Most, generally, are
scribe the system by a set of ordinary linear not. To the extent they degrade loop per-
differential equations. If you didn't know formance, the reduced perfo1·mance is
what linear meant before, likely you still simply accepted. In most cases the reduction
don't. It means that if a change is made in in perfo1·111ance would be hard to see.
the up direction, the response is exactly the
same as if made down, except reversed of So that leaves a few you cannot ignore. At
course. It means that if you double a least if you know their characteristics, you
change, you get exactly twice what can decide whether to ignore them or try to
happened before, We are aware that no •do something· to minimize their effect. I di-
physical system is linear, yet we live our vide nonlinearities into two categories,
lives and conduct our analyses as if they process and hardware. Then for the hard-
were linear. It's so much easier that way, ware I divide those nonlinearities into con-
and so often is quite acceptable. tinuous and discontinuous, the distinction
becoming more apparent as you read
further.
'
that would make a whole book in itself. Recall from Chapter 2 and Figure 2.3 that
the gain is set proportional to 1/RL.· If the
product of R times L increases, then a lower
gain should be used. If RL increases and no
change is made in gain, the loop may cycle.
Remember from Figure 4.6 and the
Hardware (and software) nonlinearities in associated discussion in Chapter 4 that L is
the continuous category are usually that way the result of combining all the smaller lags
intentionally, to compensate for some other in the loop, and the rate, R, is the response
nonlinearity. The goal is usually to make an of the longest lag in the loop.
overall system linear. A square root
extractor on a.· differential pressure Now let's look at what velocity limiting
transmitter is a prime example. With the does. A valve positioner may be able to
flexibility of computer-based controllers make a 2% change with a first order lag of
there is a wide selection of algorithms that 0.3 seconds. For a large change, say 50%, it
may be applied to make an overall system may velocity limit, · so that if you tried to
linear, or at least more nearly linear. make a first approximation to a first order
Whether a continuous hardware nonlinearity lag, the time constant might be closer to 5
is deliberate or not, the effect of it is .. · seconds. The concept is shown in Figure
analyzed in a loop much the same as a 8.2.
process nonlinearity the system is
assumed to be linear over a small operating
range, and extreme conditions are examined ;t
for potential problems. 31 Response to large change
c
0
o. Velocity llmltlng
' n~\~.11~li:~~·~e•":':•~·1·•s"'" ~·o·······•n•······•:i:1••·u••.:::
~::;=::t:',:J;L,;:;HAiL-lillhYY~.iA!•!::_ -·.'.'·::~:~U:
:· .~\:i, __ ' __ : ' ::t: :
u '. ':.o:':.:.:L.'
........ . •u<"'''•· 's·"'::\1::;:'•rn · ~:'::';:'·
• •· 31
a:
11:;·::::·• .. m··••- •:1·::t•:r·1·1·11·:~
::-:,::., ;;;;; .. .......
' " _
• ;•.• Q·Q•. IQ~!3lll.· _te~·::m• 1::::-:1•::::::::::::::: :.:::r[J••rn!!
::::::::::._:.:;;:::::,::·:;·;::.·,'" -- :·· """"""""''""'"""" ----· - -- ·--- ····- . '"'"'""'' """""""""""" -
f Response to small change
chance that th¢ gain between the controller small amount of dead band (and there can
output and the actual valve motion was not be some, even . with a positioned .valve),
one, as is assumed-, but may be only 1/2 or there will always be a .small error. The
1/4. If this is the case;'tha•the parameters R integral action keeps moving the controller
and L (from Figure 2.3) are' '' ... ~. ;ati)_\~
'• ···-·
. output until the· valve moves. The valve
determined, and so settings will not ·be . mpves. too much, eventually reversing the
accurately determined. If the tuning error. The integrating ~ti0tl..• goes to work
•
procedure is done with the controller in on that error and eventually moves the valve
• •
automatic, then the gain established for back, but again too far. Figure 8.4 shows the
small changes may be too . high for large phenomenon on a .fast. process, such as· a
changes, and the loop may become unstable. flow control loop .• On a s,lower loop the
Usually this problem, once recognized, is presence of dead band is not as easily
simply solved by reducing the gain. If high recognized, as the lag between the valve
fidelity is required ·for . small changes, movement and the process response is more
consider a positioner with higher gain. obscured.
Sometimes a booster between the positioner
and the valve motor helps, it depends on the This problem has a characteristic finger
air flow characteristics of the positioner. print. A small cycle will be seen. Sometimes
the cycle may be so small it will not be seen
If the valve is not positioned the dead band in normal records, but its effects will be
is likely to occur at larger values, like 5% or observed elsewhere in the process. Once a
10% or more. The same phenomenon just high-quality pressure transmitter on a boiler
described ·may · then occur, except steam header had a very small dead band.
everything happens at larger amplitudes. If The cycle could barely be seen on the
the valve is not positioned, then a booster pressure record but the effects of the cycle
between the controller and ·the valve isn't permeated through many users of the steam.
going help, · and the first remedial step
would be to install a positioner. Decreasing the gain, which is what is
nor1nally done if a cycle is observed, does
There is another effect of dead band that is not solve the problem. The cycle simply
probably the more important. The problem gets a longer period. If the period is
arises when using control with integral observed, normal tuning rules would call for
action, or if the process integrates. With a increasing the integral time. If this is done
Controller output
Controller output
---Actual valve motion
..
Figure 8.4. Dead band in a valve often results in a small limit cycle, characterized
by a triangular wave in the controller output and a square wave in the actual valve.
motion. Reducing the gain or increasing the integral time increases the period but
does not otherwise alter the cycle significantly.
•
the period again gets longer. If this behavior good way to check for dead band is to make
is not recognized as a dead band problem, two small changes, in manual, in the sa111e
the controller settings will be greatly direction, and then reverse· ditection with '
reduced from optimum, and the tuner will the same two small changes. ·If the process
have a high degree of frustration. does not repeat itself for the same outputs,
this is a good indication that significant ,
'
'
If the controller is in automatic then it is
harder to detect the presence of dead band, '
1
the amplitude and time scales. The positioner chosen, or a booster used
controller output will tend to be triangular between the positioner and the valve
in shape, and the actual valve motion will operator. The booster will not affect the
tend to be a square wave. Realize that this dead band but it will improve the small-
figure is idealized. What you see will not be amplitude dynamics, which often is a
that tidy. If the process is slower, what you sufficient improvement to not require any
observe may look very little like Figure 8.4. more effort. Not all positioners are created
But you will see at least the evidence of a equal in their ability to overcome dead zone.
small, relatively fixed amplitude cycle, Look for one with a high gain. The
whose period lengthens as gain is decreased manufacturer should be able to supply you
or as integral time is increased. The period with that infonnation. A gain of 50 is too
will be significantly longer than the natural low in my opinion. The ·dead band with a
period, which is approximately 4L, the L positioner is essentially the dead band
coming from Figure 2.3. without a positioner divided by the
positioner gain.
If you have a digital system with a monitor,
then the presence of dead band may be You cannot really be sure from· the control
observed when making small step changes room whether the valve is moving or not. I
in manual. These are the types of changes have used two methods. One is to simply
made in testing for process dynamics in place my fingers on the valve stem, next to
open-loop tests to deterrnine tuning settings. the packing. You will need to set up some
With small changes you may see no communication between the control room
response, or the slope, . R, will not be and the valve, to know when the controller
. proportional to the size of the step change, output has been changed, and by how much .
but rather will increase more than the size of With digital systems, you may connect a
the upset as the size is increased. Also you digital meter next to you in the field. The
may see a longer apparent dead time than human senses can ··detect very small
you might expect. If the apparent dead time changes, even less than 0.001 inches, but
gets shorter as you increase the amplitude of not quantitatively and not if they occur
the step, this is a good indicl:!-tor that you slowly. So this method ma~ be used in a
probably have significant dead band. A pinch. If you observe dead band using this
'
method, then you are sure it is there. If you limit, the positioner puts full supply or full
cannot detect it this way, then you are not vent to the valve motor. Then when the
sure whether it-is. present or not. controller signal comes back on scale, the
. .
,. -·--·-
,"';''
-,•· -
'.-.,~
.,
·-t"·'tt"'
··::«,!C;~
...,-.,
• , ,
-'
-- positiorier output has to change substantially
A better method is to .riflJM:"!'l,'Av;;:.~ · · · l?efpre th~ yalve comes off a stop. This can
micrometer to monitor the valve ~~v~1Ilerii. ·. ~·
'1'"•t'·,- '·'.-,,,. . ' - '',
linkages. It does happen. It is easy to give the advice to simply not let
. '• ·-
.. .
- -· . '
. . .
the valve get in •this predicalllent.. ·That
In summary, dead band is probably the most advice is easier to give than to follow. At
common. instrument .cause · why controller least now you can recognize it as a potential
tuning efforts' do n:ot look like the nice problem. The cycle can be stopped by going
· smooth curves presented in the articles on to manual momentarily. Sometimes it helps ·
tuning, Be
.
awar¢ofthe
. .
different ways it can to · reduce the supply pressure to the
contribute,
.
especially for small excursions. .
positioner (or controller) if the problem is at
Be particularly aware if you observe.a small that end of the scale. Be careful if you are •
amplitude cycle . whose period is limiting the controller output to avoid this
significantly longer than the loop's natural problem. It is not so critical if it is at the
period. Quite possibly the· smart valves now open end of the travel, but if it is at the
coming on. the market . will minimize the closed end you could end up not being able
problems discussed here, but there will still to close the valve. If you are using a
be non-smart valv:es ...around for several program-your-own computer/controller you
years. can write a program, using feedback from
the positioner output, to manage this
·3. Valves at limits. problem.
4. Integral Windup. own limit, far beyond the valve limit .. The
controller has to ''unwind'' and this. takes
time. Frequently this time··· is\· excessive
The phenomenon of integral (reset) windup • • • . '
has been recognized probably since. the g1v1ng rise to poor perfom1ance, and
function was invented. I believe this was in potentially to instability.
the 1930's. Initially there was little that
could be done,. and what was done was To combat this problem, instrument manu-
awkward. In the middle period of pneumatic facturers have offered a variety of solutions,
controllers (early 19SO's) instrument items generically called anti-windup· protection.
were developed to combat the probleni. Most and probably all electronic and com-
When electronic controllers came . along, puter controllers now on the market either
more elegant and simpler methods· were have the feature as standard, or offer it as an
developed. Then came digital controllers option. Pneumatic controllers generally re-
and still better ways were found. quire a marriage with external components.
A cautionary note is this all anti-windup
•
features do not perform the same way.
•
the controller will continue to change the with integral windup, it may pay to study
output (windup) based on the integral of the the details ·of how a particular vendor
error. This isn't the problem yet. · The implements the protection. It is likely that
problem occurs when the. valve needs to one vendor would be preferred to another if
come back on scale, for at. that time' tlte the windup circumstances tend to be the
.
'
'
The Ziegler and Nichols tuning rules were only the programmer knows for sure. If it is
based on the way most analog controllers PID then you should use the older algorithm
then and now are made. With the advent of to be sure the tuning rules apply. I under-
digital controllers designers and program- stand there could be certain circumstances
mers were not constrained by what could be where the newer algorithm might be pre-
economically built mechanically or elec- ferred, but I have not studied this possibil-
tronically. The result is that many, and ity. If the controller is PI (or PD, which is
perhaps most, manufacturers of digital con- not used very often) or if the controller is a
trollers offer more than one algorithm. In PLC, then I can't give you a general rule,
addition, many . laboratory-type environ- since there are so many possibilities.
ments may be using a non-instrument- Controller
vendor computer, with the control algorithm Setpoint - Output
written by their own people. It is not p I D
unusual to find these have non-classical -
algorithms combined with reasoned and Controlled
some intuitive modifications. In this case Variable
the classical tuning rules must be used with
caution, as the programmer may not have a. Series Algorithm
been knowledgeable about control· loop
dynamics and the algorithms tuning rules Controller p
have been based on. Setpoint Output
I
Various words get used to define the vari-
-
ations from the older algorithm. Sometimes
Controlled D
the older is called interacting, while vari- Variable
ations on a newer are called non-interacting.
Sometimes the older is called series and a. Parallel Algorithm
others called parallel. If the controller is
Figure 9.1. The difference between a series and a
just P or just I, then it doesn't matter, unless parallel algorithm is shown conceptually in this
the algorithm is home-grown, in which case signal-flow diagram.
- -- .
''
'
.
'
called parallel or non-interacting, the .•
P (1 + I) (1 + D)
adequate for most loops unless derivative sample and the output based on that sample. '..'
•
•
l
•
'
.,
Chapter 9, Potpourri 75
·. •·· .. . . ·.. Aliased signal 60-Hz signal
· · · •..from samples
' .~ . '
'' /
/
'
I
•
'' /
/
50-Hz samples
Figure 9.2. If a cycle is not sampled often enough it will look like a longer cycle, and create havoc.
There is another consideration if the cycle relatively fast processes, the use of
time is · long relative to the potential derivative action in a digital controller is not
variations .· in the · process, whether these recommended. It probably is not doing what
variations be noise or real. Imagine these it is supposed to do.
variations to have periodicity. Then the
sampling system . should sample several These notes are not intended to provide
times (5 or lO)>during one of these periods. adequate training in deciding when the
If it samples fewer than twice, then a sampling frequency is sufficient to avoid
potentially very· serious numerical-type aliasing, but rather are intended only to alert
problem called aliasing arises. Figure 9.1 you to the fact that sampling introduces
demonstrates the problem. problems beyond the simple · · . lags
introduced. ··
76 Chapter 9, Potpourri
is that to get the attenuation you also If your loop is of the pneumatic vintage you
introduce a lag that will affect the· might consider · · the Moore Products
apparent dead time and · hence the pneumatic ·relay, Model 59&,:,,which they•
79
0 Make the process faster, such as mixing hot and cold to control
temperature, rather than using a heat exchanger, or putting a
recirculating pump on a jacket fluid, rather than having it be just a
one-pass.
81
'
' . -._,
'- -
-
...
. ..
.
•
. . .
- . - - ·-,-;--.-
The definitions. marked below with. an asterisk (*) have been reprinted by' permission, .from
Comprehensive Dicti'9nary of Measurement and Control, 2nd Edition, by W. H. Cubberly,
copyright 1991, In$t:t ument Society of America. Some of the tertns have broad meanings, in
which case only those meanings which most apply to the use in this booklet have been copied.
Averaging level control A description of Dead band * The range through which an
the tuning method which allows the volume input may be changed without initiating an
of a tank to be used as a surge capacity, observable change in output. There · are
rather than sending upsets on to the other separate and distinct input-output relation-
parts of a process, as would happen if the ships for increasing and decreasing input
level loop were tuned by the typical tuning signals.
rules.
Derivative action * A type of control-sys-
Capacitance· The ability of a system or de- tem . action · in which a predetertnined
vice to store energy. relation exists ·between the position of the
final control element and the derivative of
Closed loop (feedback loop) See loop, the controlled variable with respect to time.
.
closed. :
: .
-
.
' -- '
84 Glossary
Gain, controller The ratio of a change in Lag * 1. A relative measure of the time
the output to a change in either the con- delay between two events, states, or mecha-
•
trolled variable or the setpoint. n1sms.
Gain, ultimate The controller gain which Limit cycle A cycle whose amplitude is
will maintain the loop in a continuous cycle limited, typically because of some nonlin-
of constant amplitude, without depending earity in the loop.
..
on nonlinearities to limit the amplitude. All . •
and derivative action must be not it effect. exists between two variables when the ratio
of the value of one variable to the corre- ·~
Gain, process Unless otherwise defmed in · sponding value of the other is constant over
context, it is the steady state ratio of the the entire range of possible values. l.,
l••
output of a process to a change in its input. •
•
CJ
~~
'
It will always have units. Load change The same as a disturbance. ••
"
.;;
Si
••
•
IAE * Integral absolute error.. A measure Loop, closed (feedback loop) * A signal
' '.j
•
•
•
of controller error defined by the integral of path which includes a forward path, a feed-
the absolute value of a time-dependent error back path and a summing point, and for111s a l
'
function; used in tuning automatic control- closed circuit. •
ler to respond properly to process transients. •
back.
IE The integral of the error. It is a measure
of the performance of a control loop to Non-self-regulating The opposite of self-
disturbances or setpoint changes, but posi- regulating.
tive errors cancel negative errors.
Open-loop unstable The same as ·condi-
ITAE * Integral time absolute error. A tionally stable.
measure of the controller error defined by
the integral of the product of time and the Period, natural The period of the cycle
absolute value of a time-dependent error present when the ultimate gain is in effect.
function; whereas the absolute value pre-
vents opposite excursions in the process Period, ultimate The same as natural
variable from canceling each other, the mul- period.
tiplication by time places a more severe
penalty on sustained transients. PID action * A mode of control action in
which proportional, integral and derivative
Integral action * A type of controller func- action are combined.
tion where the output (control) signal or
action is a time integral of the input (sensor) Proportional action That part of a control-
signal. ler's action which produces a change in the
output in proportion to a change in the con-
Integral time In a proportional...plus-inte- trolled variable.
gral controller it is the time, in an open-loop
test, for the controller, while integrating the Proportional gain See gain, controller.
error, to cause an output change equal to the
proportional change, when introducing an Proportional band * 1. The change in in-
error from an initial zero-error state. put required to produce a full range change
in output due to proportional action. It is
Inverse response * The dynamic character- reciprocally related to proportional gain.
istic of a process by which its output re-
sponds to an input change by moving in- Ramp A word used to describe a parameter
itially in one direction but finally in another. which is changing at a constant rate relative
to time.
,
-
GLOSSARY 85
Rate action * 1. Another name for the de- Self-regulating * the property of a process
rivative control mode. or machine which per111its attainment of
equilibrium , after a disturbance, without the
intervention of a controller .
Response, step * The total (transient plus Step response See response, step.
'
steady-state) time response resulting from a ·'
sudden change from one constant level of Time constant For a first-order lag, the
input to another. time to reach 63.2% of the final value in re-
sponse to a step change.
Sampling rate * For a given measurement,
the number of times it is sampled per sec- Value, desired * In process instrumenta-
ond in a time-division-multiplexed system. tion, the value of the controlled variable
Typically it is at least five times the highest wanted or chosen. The desired value equals
data frequency of the measurement. the ideal value in an idealized system.
,, •'" '•
,- ..·. •'
,,_,':.
''
·-~-- .' •, '
:-.·: - ~-
:-:-:'; ·''',_.
,
-- __
)~.
-~
-----;_-___
....-. --:-,·:~:: !il!ii'::'':·-·- :.:.:':~~'i~}-::··
,.,, '
--, '
,-,:~:.
' :>;.:.>:::·
-- _-, ,- '· - ----
--
A-2
PURE DEAD TIME PROCESS
' ,,
•
Controller
r -, Process Pure
Dead Time Controlled
I Set +
I Variable
I
KP
I Point
Kc L
I
I ..J
·.
•
~'
I I
Point I This curve is
#1 repeated from
bottom of page.
I I I
I I I
I I I
I I
I I
I I I
I I
Point
I I I I No. 2 is No. 1
#2 multiplied by
proportional gain.
I I I I (Kc= 1/2)
I I I
I l
I L P = 2L
I
I I
I I I
Point
#3 I No. 3 is #2 delayed
by L, and multiplied
I I by process gain.
(Kp = 2, arbitrarily)
I I I NoteP=2L
I I I
I I I
I I I
I
I
I
I No. 1 is the
Point I I inverse of No. 3,
#4 I for correct control
action. Note this
is -180° phase
related to No. 3
•
A-3
PROCESS WITH DEAD TIME AND INTEGRATION
Controll19r Process
r-- - --- --
Pure Integrator
1
set_
1 Kc
I Point
L,_ _ _
I I I I
bottom of page.
I I
I
I I I I
Point I I I I No. 2 is No. 1
#2 multiplied by
I proportional gain.
I I (Kc= 2)
I I I I
I I I
• I I
L
I I
I I I I
No. 3 is No. 2
Point I delayed by dead
time, L. Assume
#3
process gain for
I I I I this part = 1. Note
I I I
I 90° phase lag.
I
P = 4L No. 4 is changing
I I I I at its maximum
Point I rate when No. 3 is
#4 I I I at its maximum
I I I I and is not
I changing at all
I I I when No. 3 is
I •.· ! I
zero. Note the
I
> .
.
90° phase lag and
I I an arbitrary
I I I attenuation of 2.
I I I
I I I
I I No. 1 is the
Point .1
I I inverse of No. 4,
I I for correct control
#1 action.· Note this •
is a 180° phase
lag.
•
•
MORET TTO
;
The first edition of this boo . s self-publi•~· I didn't expect to do that but I did. It sold
close to 17,000 copies! This secon is tnott._. tvliceal -- _- and I had so much fun
publishing the first that I'm looking forward Tu this expe1·icnee. I '. ~ '~ ~ tho ex
treatment of the subject useful. With the fourth printing of the second edition the• booklet has
now sold more than 22,000 copies.
Reader Feedback
If you care to make any comments about this booklet, I would be y to hear fr01n you. And if
there is ever a second printing, I will consider incorporating your suggestions. I have left some
white space to help in that regard. I fully realize I have experienced only a part of the total field
of feedback control. c- ,
Marketing
If you think this booklet has been helpful I would appreciate your passing the word along. I am
dependent on F'RIENDLY AND INTERESTED PEOPLE to:
0
WHEN A vFitIEND GREETS YOU WITH ''WHAT'S NEW?''
YOU CAN MENTION TIDS PUBLICATION!
PUBUSHEDBY
STRAIGHT-LINE CONTROL COMPANY, INCORPORATED
3 Bridle Brook Lane
Newark, DE 197 11 -2003
(302) 731-4699, fax (302) 454-8599
E-mai l: dwstclair@aol.com
Home Page: http://members.aol.com/pidcontrol/booklet.html